Bring HNSW for fast approximate nearest Neighbour search

**Is your feature request related to a problem? Please describe.**
LLM uses cosine-similarity to retrieve relevant corpus as memory. While `KNN` is most commonly used to get `k` similar items, it's time-consuming on high-dimension vectors.

HNSW will be much efficient in finding approximate k nearest neighbor on high-dimension dataset. There's a csharp implementation we can probably leverage [HNSW](https://github.com/curiosity-ai/hnsw-sharp)

**Describe the solution you'd like**
A clear and concise description of what you want to happen.

**Describe alternatives you've considered**
A clear and concise description of any alternative solutions or features you've considered.

**Additional context**
Add any other context or screenshots about the feature request here.